解决动态约束多目标问题的复合预测进化算法

Composite predictive evolutionary algorithm for dynamic constrained multi-objective problems

中文关键词  动态多目标优化  进化算法  动态约束条件

英文关键词  dynamic multi-objective optimization  evolutionary algorithm  time-varying constraints

中文摘要

      动态约束多目标问题在路口交通管理、节能电力调度等现实场景中出现较多, 其目标函数和约束条件都会随时间(环境)发生连续缓慢变化. 求解这类动态问题的关键, 是有效追踪问题的随环境变化的一组最优解集. 为求解此类问题, 首先, 将约束变化分为2类, 并针对两类变化提出2个约束预测器, 用以追踪可行区域; 其次, 将约束预测器与非线性预测器组合成复合预测策略, 根据问题的不同变化情况使用策略中的对应预测器, 消耗较少的资源获得预测解, 加速寻优过程; 再次, 应用基于分解的多目标优化算法, 将预测解优化得到最终的最优解. 所提出的基于复合预测的动态多目标优化算法在8个动态变化的问题上与6个典型算法进行对比测试, 实验结果表明, 所提算法获得的解集在收敛性和多样性上具有显著优势, 复合预测策略的预测性能较优.

英文摘要

      Dynamic constrained multi-objective problems are used in real-world scenarios, such as intersection traffic management and energy-efficient power scheduling. Both the objective functions and constraints undergo continuous slow changes over time (in the environment). The key to solving these dynamic problems is effectively tracking a set of optimal solutions that evolve with environment. To address such problems, firstly, constraint changes are categorized into two types, and two constraint predictors are proposed for tracking feasible regions. Secondly, the constraint predictors are combined with a nonlinear predictor to form a composite predictive strategy. Depending on the specific changes in the problem, the strategy uses the corresponding predictor to obtain predictive solutions with less resource consumption, thus accelerating the optimization process. Finally, a decomposition-based multi-objective optimization algorithm is applied to optimize the predictive solutions, obtaining the ultimate optimal solution. The composite predictive evolutionary algorithms compared with six typical evolutionary algorithms on eight dynamic problems. The experimental results demonstrate that the proposed algorithm has a significant advantage in terms of convergence and diversity in the solution set, with superior predictive performance of the composite strategy.